Company:

Neklyudov's Cookbook

Role:

Developer

Period:

February 2026

Location:

Hong Kong

Company:

Neklyudov's Cookbook

Role:

Developer

Period:

February 2026

Location:

Hong Kong

before

INTERVENTION

after

Family recipes trapped in voice notes, texts, and memory.

Multimodal recipe intake, human-in-the-loop approval, website deployment.

A dynamic, searchable archive - 25+ recipes and growing.

Family recipes trapped in voice notes, texts, and memory.

Multimodal recipe intake, human-in-the-loop approval, website deployment.

A dynamic, searchable archive - 25+ recipes and growing.

The Client.

The Neklyudov Family Cookbook is an AI-powered digital repository designed to preserve generational culinary traditions. In a family of gastronomes, where recipes were historically fragmented across voice memos and unrecorded verbal traditions, this platform serves as a permanent, searchable, and interactive "Living Archive."

The Challenge.

Cultural heritage is often lost due to the high "friction" of manual documentation. The challenge was to create a zero-friction system that allows non-technical users to capture culinary data - often shared via informal voice notes or messy texts - and instantly transform it into a standardised recipe gallery.

The Solution.

I acted as the Product Architect and Developer, leveraging Cursor to rapidly build and iterate on a full-stack automation pipeline. My approach centred on building a "Human-AI Partnership" where the tech handles the logistics and the user provides the soul:

  • The "Frictionless" Intake: Users simply drop photos, voice memos, or informal texts into a Telegram Bot. This utilizes Multi-Modal Input Capture and long-polling to meet the family where they already communicate, requiring zero onboarding.

  • Structured Data Schema: OpenAI’s Whisper was chosen for its high-accuracy voice-to-text transcription. GPT-4o (Chat Completions) with strict JSON output, allows the system to parse unstructured input into a standardized format (ingredients, steps, macros, and tags), which is then stored in a PostgreSQL database (Supabase).

  • The Human-in-the-Loop Interface: Rather than relying on "black-box" automation, the bot engages in an interactive dialogue. It presents a preview for user verification, allowing for manual adjustments, precise measurements, or additional "behind-the-scenes" photos. This ensures the AI serves as a "copilot" that respects human nuance.

  • Web Deployment: Once the user "approves" the draft, the data is pushed to a responsive Next.js gallery, utilizing Server-Side Rendering (SSR) to create a living, searchable archive that feels like a premium digital publication.

The Impact.

1 Familial Heritage Website

Successfully converted "messy," fragmented data into a structured, searchable, and permanent digital legacy.

22 Recipes and Growing

Provided a scalable proof-of-concept for how modern tech stacks can solve legacy data problems in a human-centric way.

Does this spark an idea?

Let's bring it to life together!

Does this spark an idea?

Let's bring it to life together!

Discover Similar Projects.

Discover Projects.